Robert,
All this discussion is unfortunately moot, since it depends entirely
on how the computer is programmed and what kinds of sensory inputs it
is hooked up to, and what kind of control over active devices with
feedback it is given. There is near infinite possible variation here.
So one can trivially imagine computers of any type with any set of
capabilities. To see this just consider my contention is that if a
computer were constructed out of biological materials according to a
human genetic blueprint then it would be indistinguishable from a
human. Therefore the question of whether computers are sentient or
conscious or not depends entirely on the definition of consciousness
and the design of the computer. It is trivially simple whether
computers are/can be conscious or not. The answer is YES. One simply
defines consciousness and then one builds the computer to those
specs. For the very conservative that might mean building one to
human specs (what age level?) but it is always theoretically possible.
In my view all computers are conscious of what they are conscious of,
as are all organisms. One must never fall into the fallacy of
confusing self-consciousness with consciousness. That is just
consciousness of the concept of a self which is only one of the many
contents of consciousness.
Edgar
On Aug 24, 2008, at 5:18 AM, Robert Karl Stonjek wrote:
----- Original Message -----
From: Stan Franklin
To: Evolutionary Psychology list
Sent: Sunday, August 24, 2008 5:36 AM
Subject: Fwd: [evol-psych] Essay: The Shy Computer
Robert, I enjoyed your The Shy Computer essay. Thanks for it. My
comments are below. All the best. Stan
Our first observation is that, in response to stimuli or
instruction, the robot can indeed perform many cognitive
functions. But when the response to stimuli is complete it more or
less stops. Its curiosity appears to be largely absent and the
only conversation it engages in revolves around its basic needs (as
programmed in to simulate the living - eg hunger, thirst).
Curiosity (exploration) is often built in to software agents and
even some robots as being necessary for learning. Some autonomous
software agents and robots generate their own internal stimuli to
which they respond. They have their own agenda. My IDA agent would
sometime initiate correspondence with sailor about new jobs when
the sailor had not contacted IDA in a timely fashion.
... how does one program in a general curiosity without telling the
robot what to be curious about?
Every autonomous agent must have some built-in primitive sensing,
motivating, and acting capabilities. Built-in curiosity may simply
be the motivation to do, or to pay attention to, the novel. This,
of course, requires memory of an appropriate type, which can also
be built-in.
But the robot takes no interest in the drawings it has made
previously nor those made by other people.
It certainly can, depending on the built-in motivations.
RKS:
You would have to specify each domain of curiosity quite
specifically. This is not human-like. Humans are continuously
finding new areas of curiosity. I was considering the computer as
an analogue of human consciousness or human evolved behaviour
rather than the idea of making a software agent.
Sure you can manually program a software agent to be curious in a
particular way in a particular domain, but is this how humans do it?
I would also point out that when humans need extra information to
complete a task they seek it, but that is not the same as curiosity
by the common useage of the word. Curiosity is the seeking of
information for no obvious or immediate reason.
The Robot talks almost endlessly, but mainly not to anyone - it
just talks. And it doesn't take much interest in the chatter of
others.
There's no reason the robot shouldn't speak only when it seems
appropriate to the robot, nor that it shouldn't be interested in
what others say.
RKS:
If expression (output) were programmed in the way I have suggested
then a robot would start acting like a toddler - they talk
endlessly, but not to anyone (90% of verbal output occurs without a
listener present). This is what I am trying to understand - what
would make a robot start acting like a child?
Thus three processes are in play - expression, curiosity, and the
processing of information into a single composite (self) image. As
both the internal and external consciousness can sense both the
world and the physical self and each other, the stage is set for
all the complexity normally found in humans.
Expression, in the form of email messages in English, and the
processing of both internal and external stimuli are both a part of
our IDA software agent. To set the stage for human-like complexity,
more is needed, namely learning in several modes. This led us to
LIDA, Learning IDA.
RKS:
Why does it learn? Because you have specifically instructed it to
do so. This form of prescriptive instruction is not allowed in my
model, as I mentioned.
A robot designed with human-like modules would not be anything like
IDA or her sister LIDA. A human-like robot would have to sleep or
would stop functioning normally after a few days (for instance).
Note that your software agent always does as you have instructed it
to do. Does one, after learning, want to take on some other task?
Does it want to become a music composition agent instead of filling
out forms for the military?
This uncertainty, which we would find in the model I have
suggested, means that the robot in my piece is an analogue of human
consciousness and so could not be reliably set up for one
particular task.
Note also that the more effective the information gathering skills
of the agent are, the less information you need to supply
initially. For instance, a pony can walk just hours after being
born, can recognise the mother and feed by itself. Humans can not,
but humans are born with far less innate knowledge - we are more
effective information gatherers and processors.
Thus if your software agent was to evolve toward human-ness you
will find that the amount of prescriptive information required
falls off as it evolves but the number of personalities,
predispositions and career options balloons considerably until the
only way that you could reliably create an agent for a specific
task is to select from a population of agents those that are
'naturally' predisposed to the proposed career.
Kind Regards
Robert Karl Stonjek
--
Stan Franklin Professor Computer Science
W. Harry Feinstone Interdisciplinary Research Professor
Institute for Intelligent Systems
FedEx Institute of Technology
The University of Memphis
Memphis, TN 38152 USA
phone 901-678-1341
fax 901-678-5129 [EMAIL PROTECTED]
www.cs.memphis.edu/~franklin
